Data Mining MCQs

In high dimensions, data become sparse and distances often become less informative.
Algorithms may require much more data and careful feature selection.

Redundancy can waste storage and overweight repeated information.
It may also create multicollinearity or duplicate counting.

Pearson correlation measures linear dependence only.
Curved relationships can be strong while producing a near-zero coefficient.

Lift above one means Y is more common when X occurs than in the dataset overall.
The rule reflects positive dependence.

High support reflects frequent co-occurrence in absolute terms.
Lift near one shows the co-occurrence is close to what marginal popularity predicts.

Hash trees organize candidates into buckets for efficient matching.
Transactions can be mapped to relevant candidate subsets during scans.

Interestingness combines statistical strength with practical relevance.
A frequent pattern may still be unhelpful if it is obvious or unactionable.

Reduction can lower storage, computation, and noise.
Examples include sampling, aggregation, feature selection, and dimensionality reduction.

Constructed features may expose useful relationships not directly present in raw variables.
Examples include ratios, interactions, and time-based indicators.

A lift of one means confidence equals the baseline support of the consequent.
Knowing X provides no association-based advantage for predicting Y.

Lift compares rule confidence with the consequent's baseline support.
A value above one indicates improvement over the baseline frequency.

As itemset size grows, short or nonmatching transactions cannot contribute support.
Removing them reduces later scan cost.